Freezing is a hallmark of fear memory retrieval, yet it does not represent a uniform physiological state. We have shown that rats display two distinct forms of freezing, characterized by different oscillatory profiles in the basolateral amygdala (BLA) and medial prefrontal cortex (mPFC). Freezing episodes enriched in 4-Hz activity were most common during early retrieval, showing increased power in this band with suppression of theta, consistent with fear expression. By contrast, theta-enriched freezing emerged more often as extinction progressed, with stronger theta and reduced 4-Hz power, suggesting a distinct and possibly transitional state linked to fear inhibition. Although indistinguishable in terms of immobility, both states showed robust and opposing signatures in oscillatory activity, neuronal firing, and interregional interactions across BLA and mPFC, indicating that a similar outward behavior can arise from distinct internal dynamics. In the present study we extend these findings by exploring whether such subtypes can be identified without relying on arbitrary cutoffs. We applied dimensionality reduction and clustering analysis to behavioral and neural variables to test whether freezing can be decomposed into functionally meaningful sub-states in an unsupervised manner. This approach provides a new framework to examine how fear memories are retrieved and extinguished through the study of divergent freezing behaviors reflecting distinct internal states.